confluent persistence, on data. which changes how you think about concurrency.
it allows you to "branch, work on a worktree, and merge" but unlike git worktrees you don't need to copy a complete clone of the data, you instead, like in the git object store, share most of it and only need memory for what you touch in your branch.
and your data only changes when _you_ decide to rebase, and merging partial work results is a _deliberate_ step, not something happening under your feet.